Q. How do I prepare tensor?
A. Both BIGtensor (Single) and BIGtensor (Distributed) work on graphs in TAB-separated plain text format.
Each line corresponds to an element, and contains indices of each mode and the value.
The index starts from 1.
For example, here is a 3-mode tensor.
The first line shows that the element at the position (1, 2, 4) is 1,
and the second line shows that the element at the position (3, 4, 1) is 2.
1 2 4 1 3 4 1 2 4 1 3 1 1 1 1 2 1 3 3 1 1 4 1 2 1 1 2 1 1 2 1 2
Q. How do I setup a machine for BIGtensor (Single)?
BIGtensor (Single) is running on any Linux based machine including a c++ compiler with OpenCL, CUDA environment. Please refer to the following document for detail:
Q. How do I setup hadoop cluster for BIGtensor (Distributed)?
A. In most cases, you will use a Hadoop cluster set up by someone else, and thus you don't need to set up a Hadoop cluster by yourself. However, you might want to set up a Hadoop cluster for yourself for some occasions. In that case, here are several useful documents:
Hadoop Cluster Setup
Cloudera Hadoop Distribution
Q. What are the main advantages of BIGtensor?
A. First, the advantage of BIGtensor (Single) is the flexibility. Depending on your machine environment, you can select a CPU or GPU based methods to accelerate the tensor analysis computations. Second, the most important advantage of BIGtensor (Distributed) is the scalability. BIGtensor (Distributed) provides tensor mining algorithms for billion-scale tensors which are several orders of magnitude larger than previous works can handle. Other advantages include the ability to analyze tensor data on Hadoop clusters.
Q. What is the future of BIGtensor?
A. BIGtensor will be extended to include more tensor analysis algorithms, and incorporate efficient indexing methods for tensors.